The Effect of Using Mobile Technology-Based Methods That Record Food or Nutrient Intake on Diabetes Control and Nutrition Outcomes: A Systematic Review

(1) Background: Mobile technologies may be utilised for dietary intake assessment for people with diabetes. The published literature was systematically reviewed to determine the effect of using mobile electronic devices to record food or nutrient intake on diabetes control and nutrition outcomes; (2) Methods: The review protocol was registered with PROSPERO: registration number CRD42016050079, and followed PRISMA guidelines. Original research of mobile electronic devices where food or nutrient intake was recorded in people with diabetes with any treatment regimen, and where this intervention was compared with usual care or alternative treatment models, was considered. Quality was assessed using the Quality Criteria Checklist for Primary Research; (3) Results: Nine papers formed the final library with a range of interventions and control practices investigated. The food/nutrient intake recording component of the intervention and patient engagement with the technology was not well described. When assessed for quality, three studies rated positive, five were neutral and one negative. There was significantly greater improvement in HbA1c in the intervention group compared to the control group in four of the nine studies; (4) Conclusion: Based on the available evidence there are no clear recommendations for using technology to record dietary data in this population.


Introduction
The use of mobile technology in everyday life continues to increase exponentially. By 2019, the number of smartphone users worldwide is expected to grow to 2.7 billion, while there will be 1.4 billion tablet users [1,2]. Rates of diabetes in China (11.6%) [3] have now overtaken those reported in the United States (9.3%) [4] with diabetes considered to be one of the most challenging health problems of the 21st century [5]. There are vast opportunities for interventions for diabetes that utilise or are delivered through mobile phones or tablets (i.e., mobile applications) or other portable devices.
Recommendations for optimal management of diabetes have been outlined by the American Diabetes Association [6]. There should be organised and coordinated approaches with collaboration between the health care team and the patient who has an active role in their self-management. Measuring, monitoring, analysing, communicating and acting on a range of parameters, in addition to blood glucose levels, are integral to effective management. Mobile technologies provide an interface for this to occur, and there is some evidence of their effectiveness. For example, a systematic review identified statistically significant and clinically relevant declines in HbA1c levels for adults receiving telemedicine applications with personalised feedback compared to non-telemedicine treatment approaches [7]. Technology has also been shown to be preferential to weighed food records for recording dietary intake information in people with type 2 diabetes [8].
A recent systematic review [9] evaluated 65 free to download English language applications for the self-management of diabetes. Whilst there was a large number (956 applications across Google Play, App Store and Windows Phone Store) and wide selection of applications available, reviewers identified that the versatility and inclusion of features was highly variable. Only eight of the applications reviewed included a food database with energy, carbohydrate or fluid intake input, with an average time for meal log of entry of less than one minute to five minutes [9].
The effect of mobile devices where food or nutrition intake is recorded has been considered in other chronic conditions including weight management [10] and chronic renal disease [11]. These reviews reported varying levels of effect, whilst opportunities for further development of applications were identified. No similar review has been undertaken of studies comparing the use of mobile devices for recording nutritional intake compared with usual care in diabetes management. We aimed to systematically review the published literature to determine the effect of using mobile electronic devices to record food or nutrient intake on diabetes control and nutrition outcomes.

Methods
The review protocol was registered prospectively with PROSPERO, registration number CRD42016050079. In a deviation to the registered protocol, interventions were extended from mobile applications only (mobile phone or tablet) to include mobile electronic devices (including stand-alone portable devices such as personal digital assistants (PDA)). The PRISMA statement [12] was followed throughout all stages of the review.

Eligibility Criteria
The PICO (Participant-Intervention-Comparator-Outcomes design) format of Shamseer et al. [13] was used to develop criteria for review inclusion. Original research among people with type 1 or 2 diabetes mellitus or gestational diabetes (excluding pre-diabetes or diabetes prevention) with any treatment regimen, using mobile electronic devices where food or nutrient intake was recorded (alone or in addition to other parameters) and compared with usual care or alternative treatment models was considered. No age or gender restrictions were applied. Interventions consisting of text messages, phone calls, and access to internet or websites only were ineligible, although these were acceptable if delivered in addition to the intervention of interest described above. Studies with no intervention (e.g., cross-sectional studies) or no control group (e.g., before-and-after studies), reviews, opinions or commentaries, protocol papers, conference abstracts, book chapters and case reports were excluded from the review.
HbA1c was the primary outcome due to its value in assessing treatment effectiveness in patients with diabetes. It reflects glycemic control over a 3 months or 90 days average plasma glucose concentration whilst being correlated with microvascular and neuropathic complications [6]. Secondary outcomes were: other nutritional outcomes (blood glucose levels, anthropometric measures and lipids), participant engagement and adherence with nutritional data entry, and satisfaction and feasibility of nutritional data entry.

Search Terms
Search terms with a focus on the population and intervention were determined through the examination of key words used in the relevant literature. The search strategy used is shown in Figure 1. No limits were applied.

Study Selection
Six databases, Ovid MEDLINE, Embase, CINAHL, EBM Reviews-Cochrane Database of Systematic Reviews, PsycINFO and EBM Reviews-Health Technology Assessment, were searched to identify publications of relevance from date of commencement to September 2016. Reference lists of papers included in the final library were also reviewed to identify additional studies for inclusion.
The process of identification, screening and eligibility assessment was applied to ensure that all relevant studies were included. After duplicates were removed, two authors independently screened titles and abstracts, then full text publications (J.P., J.C.). Where conflicting opinions arose, these were resolved through consensus.

Data Extraction and Quality Assessment
A template was developed to extract relevant data from the original papers with data extraction completed by one author (J.C.). Two authors independently rated study quality using the Quality Criteria Checklist for Primary Research (J.P., C.E.H.) [14]. This tool considers aspects of dietary measurement and error, and is specific for studies in nutrition and dietetics. A rating of negative (weak quality, generalisability, data collection and analysis, likely bias), neutral (neither exceptionally strong nor exceptionally weak quality) or positive (strong quality, generalisability, data collection and analysis, limited bias) was assigned for each study using the rating scale provided by the checklist. Where details were not provided or authors considered an "unsure" response for specific aspects within each validity question, a final response of "no" was made.

Synthesis of Results
Results were described narratively. Due to the large variance in the types and length of interventions (including the input of nutrition data), authors considered that a meta-analysis was inappropriate.

Results
The study selection process is illustrated in Figure 2. The database search yielded 2705 records. The titles and abstracts were screened for 1687 records, with full text reviewed for 17 manuscripts. Of these, six studies were included [15][16][17][18][19][20] in addition to one study [21] known to authors and two studies [22,23] identified from hand searching of reference lists, with a final library of nine studies [15][16][17][18][19][20][21][22][23]. No previous reviews of the research question were identified.

Study Selection
Six databases, Ovid MEDLINE, Embase, CINAHL, EBM Reviews-Cochrane Database of Systematic Reviews, PsycINFO and EBM Reviews-Health Technology Assessment, were searched to identify publications of relevance from date of commencement to September 2016. Reference lists of papers included in the final library were also reviewed to identify additional studies for inclusion.
The process of identification, screening and eligibility assessment was applied to ensure that all relevant studies were included. After duplicates were removed, two authors independently screened titles and abstracts, then full text publications (J.P., J.C.). Where conflicting opinions arose, these were resolved through consensus.

Data Extraction and Quality Assessment
A template was developed to extract relevant data from the original papers with data extraction completed by one author (J.C.). Two authors independently rated study quality using the Quality Criteria Checklist for Primary Research (J.P., C.E.H.) [14]. This tool considers aspects of dietary measurement and error, and is specific for studies in nutrition and dietetics. A rating of negative (weak quality, generalisability, data collection and analysis, likely bias), neutral (neither exceptionally strong nor exceptionally weak quality) or positive (strong quality, generalisability, data collection and analysis, limited bias) was assigned for each study using the rating scale provided by the checklist. Where details were not provided or authors considered an "unsure" response for specific aspects within each validity question, a final response of "no" was made.

Synthesis of Results
Results were described narratively. Due to the large variance in the types and length of interventions (including the input of nutrition data), authors considered that a meta-analysis was inappropriate.
In all included studies the intervention was a multi-component diabetes management strategy, where dietary data was recorded in addition to a range of other medical information (e.g., blood glucose levels, medications, physical activity). In some studies, only carbohydrate intake [15,16,18,20,22] was recorded and in others food/meal intake [17,19,21,23] was collected. Interventions were delivered via a mobile phone applications for three studies [15,20,21] and as a 'system' or 'software' accessible via mobile phone for four studies [16,18,19,22]. One study utilised a PDA [17] and another [23] a purpose-developed hand held device. Some studies included additional components as part of the intervention.
Eight of the studies were randomised controlled trials (RCT) [15][16][17][18][19][20][21][22] whilst one [23] was of crossover design. In some interventions the dietary input was analysed by a database or a clinician and made available to the patient [19,23] but this was unclear or not reported in other studies. Overall, the nutrition component of the application was poorly described in most studies with a lack of information on what information was captured, whether items were selected from a database of foods and if nutrient analysis was provided to the patient. Participants were adults with type 1 diabetes [15,16,22], type 2 diabetes [17][18][19], or either condition [20]. Tsang et al. [23] did not report the participant group. No studies undertaken in women with gestational diabetes were identified. Studies were identified from Europe [15,16,21,22], the United States [17,18] and Asia [19,20,23]. Study characteristics are described in Table 1.
In all included studies the intervention was a multi-component diabetes management strategy, where dietary data was recorded in addition to a range of other medical information (e.g., blood glucose levels, medications, physical activity). In some studies, only carbohydrate intake [15,16,18,20,22] was recorded and in others food/meal intake [17,19,21,23] was collected. Interventions were delivered via a mobile phone applications for three studies [15,20,21] and as a 'system' or 'software' accessible via mobile phone for four studies [16,18,19,22]. One study utilised a PDA [17] and another [23] a purpose-developed hand held device. Some studies included additional components as part of the intervention.
Eight of the studies were randomised controlled trials (RCT) [15][16][17][18][19][20][21][22] whilst one [23] was of cross-over design. In some interventions the dietary input was analysed by a database or a clinician and made available to the patient [19,23] but this was unclear or not reported in other studies. Overall, the nutrition component of the application was poorly described in most studies with a lack of information on what information was captured, whether items were selected from a database of foods and if nutrient analysis was provided to the patient.  The assessment of quality across the included library yielded variable results (Table 2). Three studies rated positive [16,17,21], five neutral [15,[18][19][20]22] whilst one negative [23] due to a range of methodological and reporting concerns. Two particular issues were consistently noted. Bias was introduced through the participant selection process in many studies whereby only those who owned or had internet access or could reportedly use the technological device were eligible for inclusion. Although this may have enhanced study completion rates, the opportunity to participate was effectively only offered to a subsample of those with diabetes. This type of selection bias may make the intervention more relevant to the population being studied as participants were already engaged in mobile technology; this cannot be determined without understanding the reasons why some people with diabetes are not engaged in the use of technology. The statistical analysis in several cases were not reproducible, particularly the intention to treat analysis of RCTs. Table 2. Quality assessment of studies comparing the effect of using mobile devices to record food or nutrition intake to usual care on diabetes control and nutrition outcomes.

Author/Year
Quality Rating a

Validity Items b Example of Reasons for Downgrading 1 2 3 4 5 6 7 8 9 10
Drion et al., 2015 [15] Neutral Methods used in the intention to treat analysis not described Zhou et al., 2016 [20] Neutral √ Prospective participants were excluded if they were unable to use a smartphone. Statistical analysis was unclear √ = response of "yes" to the validity question; x = response of "no" to the validity question; a Assessed using The Quality Criteria Checklist for Primary Research [14]; b Validity items: [1] research question stated; [2] subject selection free from bias; [3] comparable study groups; [4] method for withdrawals described; [5] blinding used; [6] interventions described; [7] outcomes stated, measurements valid and reliable; [8] appropriate statistical analysis; [9] appropriate conclusions, limitations described; [10] funding and sponsorship free from bias. Validity items 2, 3, 6, 7 must be satisfied for a positive quality rating.
There was a statistically significantly greater improvement in HbA1c in the intervention group compared to the control group in four of nine studies [18][19][20]23]. Due to the multiple and varied components of the intervention and usual care, it was not possible to attribute whether the effect (or lack of) on HbA1c was attributable to recording of food or nutrient intake using a mobile device.
Secondary outcome data are reported in Table A1. Two studies [19,20] reported significant improvements in fasting blood glucose in favour of the intervention, alongside reductions in HbA1c. There were no significant improvements in other nutritional outcomes associated with the intervention. One study [22] reported a significantly greater reduction in triglycerides (but not other blood lipids) among patients with type 1 diabetes using mobile software with capacity to record blood glucose and carbohydrate. Patients' engagement in the food and nutrient recording aspect of the intervention was not reported in all but two studies. Waki et al. [19] reported that 40% of participants recorded dietary intake data, with a drop off observed over time. In the study by Tsang et al. [23], the frequency of dietary data transmission by participants to obtain nutrient analysis was highly variable, while the majority (73%) of participants transmitted data for three meals at once. Where assessed [15,19,20,23], patients were satisfied with the intervention, but this did not specifically relate to the food or nutrient recording component.

Discussion
Given the available number of diabetes applications, the size of the included library evaluating the effect of using mobile electronic devices to record food or nutrient intake on diabetes control and nutrition outcomes was smaller than anticipated. Several studies rated strongly from a quality perspective, although bias was introduced through participant selection and lack of methodological description.
Some studies did report favourable effects of the intervention on HbA1c however these were not in the majority. All interventions captured multiple pieces of information (e.g., blood glucose, physical activity) in addition to dietary data. Some interventions also included other components such as interaction with health professionals. Consequently, in the studies where benefits were seen, it was not possible to attribute this to the recording of dietary information alone. Mobile applications that record food and nutrient intake data only (e.g., Easy Diet Diary) are available, but their effect is unknown as they were not utilised in the studies included in this review. It is notable that in all four studies where there were significant benefits for HbA1c, the intervention involved providing the patient with analysis and/or feedback from a clinician on data captured. It is possible that recording of diabetes related information alone is insufficient to affect outcomes with benefits instead arising from multifactorial diabetes management.
Furthermore, specific detail relating to the nutritional component of the interventions, or the engagement and update by participants was not well described. This was not anticipated, but may be explained by the fact that most interventions were multi-component and did not intend to evaluate this aspect specifically. Evidently for there to be an effect, patients must comply with recording of dietary data, and reporting on adherence is essential. In most studies, inclusion criteria were having or being familiar with a mobile phone. While this excluded a proportion of patients with diabetes, it means that participants recruited were an appropriate target group for a technological based intervention.
Subsequently, we are unable to make draw conclusions as to whether using mobile devices to record food or nutrient intake affects clinical outcomes, or give recommendations as for the development or enhancement of the nutritional models used within diabetes applications. It appears that there is no harm in using mobile devices to record dietary information among patients who are familiar with mobile phone technology.
The results of this review contrasts with other reviews of the effectiveness of nutrition focused mobile applications on clinical outcomes in chronic disease populations. In overweight and obese populations with cardiovascular risk factors, Stephens et al. [10] identified beneficial impacts of text messaging or smart phone applications for reducing physical inactivity and/or overweight/obesity. A review in patients with chronic renal failure [11] found potential for clinical benefits, but no significant changes in nutritional outcomes (energy, protein, potassium, phosphorus or fluid intake, and intradialytic weight gain) through interventions of mobile application.
Some limitations at the study and review level were noted. The choice of HbA1c as a primary outcome may have limitations in several clinical conditions that affect HbA1c readings (e.g., certain anemias and in situations of abnormal red cell turnover) [6]. Consistent with the systematic review of telemedicine in the diabetes population [7], blinded outcome assessment was not commonly implemented. However the use of objective biochemical measures provides low risk of bias in their assessment. Conversely, strengths to the review include that there were no date or language restrictions applied, and the search strategy was wide reaching through the use of broad search terms and multiple databases.
There are opportunities for clinicians to engage with application developers to develop models for clinical trials. Maintaining the currency of the mobile devices (and applications) presents a challenge to program developers. Two studies in this review utilised PDAs/purpose developed handheld devices which have been now been superseded by smartphones and tablets. Carrying a second device for diabetes monitoring in addition to a mobile phone would likely be considered cumbersome. Future studies should report specific details relating to uptake and engagement of the participant with the intervention, the application (including process of nutrition data entry and use of a database) and the involvement and role of clinicians to enable reproducibility and comparison with other applications and studies. The use of HbA1c as a primary outcome would enable comparison between other studies that were limited to reporting of glucose monitoring in the absence of nutrition intake assessment. In order to fully understand the impact on diabetes of self-monitoring of diet using mobile devices, large RCTs are required comparing this intervention with a control group where dietary information is recorded using traditional methods (i.e., paper based diary) or not recorded at all.

Conclusions
Although technology may offer novel solutions to support measurement of dietary intake and improve clinical outcomes in people with diabetes, based on the present evidence, we are unable to define clear recommendations for nutrition technology use in this population.
Acknowledgments: This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
Author Contributions: All authors conceived and designed the review; J.P. completed the study selection, quality assessment, interpreted the data, and wrote the manuscript; C.E.H. completed quality assessment; H.T. provided oversight; J.C. completed the literature search, study selection, and completed data extraction. All authors contributed to preparation and critical review of the manuscript and have agreed to its submission for publication.

Conflicts of Interest:
The authors declare no conflict of interest. Table A1. Secondary outcomes of studies exploring the effect of using mobile devices to record food or nutrition intake on diabetes control and nutrition outcomes.  Significant reduction within both groups, with significant difference in change between groups in favour of the intervention (p < 0.01).
No change in weight, BMI or waist circumference within either group, and no difference in change between groups (p not reported).
No change in LDL within either group and no difference in change between groups (p not reported).
-84% of patients in the intervention group were satisfied with the app.
Rossi et al., 2010 [22] No change within either group, and no difference in change between groups (p = 0.13).
Significant increase in weight within the control group but no change within the intervention group.
No difference in change in weight between groups (p = 0.22).
No change within either group for TAG, but significant difference in change between groups in favour of the intervention (p = 0.04). No change within either group and no difference in change between groups for total cholesterol (p = 0.33) or LDL (p = 0.79). Significant increase in HDL within control group but no change within intervention group and no difference in change between groups (p = 0.14).